blob: 8bc15b362d239f211ca0e28a1054df9b2094804c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namespace.c
3 *
4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
6 *
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
10
11#include <linux/config.h>
12#include <linux/syscalls.h>
13#include <linux/slab.h>
14#include <linux/sched.h>
15#include <linux/smp_lock.h>
16#include <linux/init.h>
17#include <linux/quotaops.h>
18#include <linux/acct.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080019#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/module.h>
21#include <linux/seq_file.h>
22#include <linux/namespace.h>
23#include <linux/namei.h>
24#include <linux/security.h>
25#include <linux/mount.h>
26#include <asm/uaccess.h>
27#include <asm/unistd.h>
Ram Pai07b20882005-11-07 17:19:07 -050028#include "pnode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
30extern int __init init_rootfs(void);
31
32#ifdef CONFIG_SYSFS
33extern int __init sysfs_init(void);
34#else
35static inline int sysfs_init(void)
36{
37 return 0;
38}
39#endif
40
41/* spinlock for vfsmount related operations, inplace of dcache_lock */
Al Viro5addc5d2005-11-07 17:15:49 -050042__cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
43
44static int event;
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
46static struct list_head *mount_hashtable;
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070047static int hash_mask __read_mostly, hash_bits __read_mostly;
Ram Paib58fed82005-11-07 17:16:09 -050048static kmem_cache_t *mnt_cache;
Ram Pai390c6842005-11-07 17:17:51 -050049static struct rw_semaphore namespace_sem;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
51static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
52{
Ram Paib58fed82005-11-07 17:16:09 -050053 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
54 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 tmp = tmp + (tmp >> hash_bits);
56 return tmp & hash_mask;
57}
58
59struct vfsmount *alloc_vfsmnt(const char *name)
60{
Ram Paib58fed82005-11-07 17:16:09 -050061 struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 if (mnt) {
63 memset(mnt, 0, sizeof(struct vfsmount));
Ram Paib58fed82005-11-07 17:16:09 -050064 atomic_set(&mnt->mnt_count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 INIT_LIST_HEAD(&mnt->mnt_hash);
66 INIT_LIST_HEAD(&mnt->mnt_child);
67 INIT_LIST_HEAD(&mnt->mnt_mounts);
68 INIT_LIST_HEAD(&mnt->mnt_list);
Miklos Szeredi55e700b2005-07-07 17:57:30 -070069 INIT_LIST_HEAD(&mnt->mnt_expire);
Ram Pai03e06e62005-11-07 17:19:33 -050070 INIT_LIST_HEAD(&mnt->mnt_share);
Ram Paia58b0eb2005-11-07 17:20:48 -050071 INIT_LIST_HEAD(&mnt->mnt_slave_list);
72 INIT_LIST_HEAD(&mnt->mnt_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 if (name) {
Ram Paib58fed82005-11-07 17:16:09 -050074 int size = strlen(name) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 char *newname = kmalloc(size, GFP_KERNEL);
76 if (newname) {
77 memcpy(newname, name, size);
78 mnt->mnt_devname = newname;
79 }
80 }
81 }
82 return mnt;
83}
84
85void free_vfsmnt(struct vfsmount *mnt)
86{
87 kfree(mnt->mnt_devname);
88 kmem_cache_free(mnt_cache, mnt);
89}
90
91/*
Ram Paia05964f2005-11-07 17:20:17 -050092 * find the first or last mount at @dentry on vfsmount @mnt depending on
93 * @dir. If @dir is set return the first mount else return the last mount.
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 */
Ram Paia05964f2005-11-07 17:20:17 -050095struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
96 int dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -070097{
Ram Paib58fed82005-11-07 17:16:09 -050098 struct list_head *head = mount_hashtable + hash(mnt, dentry);
99 struct list_head *tmp = head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 struct vfsmount *p, *found = NULL;
101
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 for (;;) {
Ram Paia05964f2005-11-07 17:20:17 -0500103 tmp = dir ? tmp->next : tmp->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 p = NULL;
105 if (tmp == head)
106 break;
107 p = list_entry(tmp, struct vfsmount, mnt_hash);
108 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
Ram Paia05964f2005-11-07 17:20:17 -0500109 found = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 break;
111 }
112 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 return found;
114}
115
Ram Paia05964f2005-11-07 17:20:17 -0500116/*
117 * lookup_mnt increments the ref count before returning
118 * the vfsmount struct.
119 */
120struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
121{
122 struct vfsmount *child_mnt;
123 spin_lock(&vfsmount_lock);
124 if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
125 mntget(child_mnt);
126 spin_unlock(&vfsmount_lock);
127 return child_mnt;
128}
129
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130static inline int check_mnt(struct vfsmount *mnt)
131{
132 return mnt->mnt_namespace == current->namespace;
133}
134
Al Viro5addc5d2005-11-07 17:15:49 -0500135static void touch_namespace(struct namespace *ns)
136{
137 if (ns) {
138 ns->event = ++event;
139 wake_up_interruptible(&ns->poll);
140 }
141}
142
143static void __touch_namespace(struct namespace *ns)
144{
145 if (ns && ns->event != event) {
146 ns->event = event;
147 wake_up_interruptible(&ns->poll);
148 }
149}
150
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
152{
153 old_nd->dentry = mnt->mnt_mountpoint;
154 old_nd->mnt = mnt->mnt_parent;
155 mnt->mnt_parent = mnt;
156 mnt->mnt_mountpoint = mnt->mnt_root;
157 list_del_init(&mnt->mnt_child);
158 list_del_init(&mnt->mnt_hash);
159 old_nd->dentry->d_mounted--;
160}
161
Ram Paib90fa9a2005-11-07 17:19:50 -0500162void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
163 struct vfsmount *child_mnt)
164{
165 child_mnt->mnt_parent = mntget(mnt);
166 child_mnt->mnt_mountpoint = dget(dentry);
167 dentry->d_mounted++;
168}
169
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
171{
Ram Paib90fa9a2005-11-07 17:19:50 -0500172 mnt_set_mountpoint(nd->mnt, nd->dentry, mnt);
173 list_add_tail(&mnt->mnt_hash, mount_hashtable +
174 hash(nd->mnt, nd->dentry));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500176}
177
178/*
179 * the caller must hold vfsmount_lock
180 */
181static void commit_tree(struct vfsmount *mnt)
182{
183 struct vfsmount *parent = mnt->mnt_parent;
184 struct vfsmount *m;
185 LIST_HEAD(head);
186 struct namespace *n = parent->mnt_namespace;
187
188 BUG_ON(parent == mnt);
189
190 list_add_tail(&head, &mnt->mnt_list);
191 list_for_each_entry(m, &head, mnt_list)
192 m->mnt_namespace = n;
193 list_splice(&head, n->list.prev);
194
195 list_add_tail(&mnt->mnt_hash, mount_hashtable +
196 hash(parent, mnt->mnt_mountpoint));
197 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
198 touch_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199}
200
201static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
202{
203 struct list_head *next = p->mnt_mounts.next;
204 if (next == &p->mnt_mounts) {
205 while (1) {
206 if (p == root)
207 return NULL;
208 next = p->mnt_child.next;
209 if (next != &p->mnt_parent->mnt_mounts)
210 break;
211 p = p->mnt_parent;
212 }
213 }
214 return list_entry(next, struct vfsmount, mnt_child);
215}
216
Ram Pai9676f0c2005-11-07 17:21:20 -0500217static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
218{
219 struct list_head *prev = p->mnt_mounts.prev;
220 while (prev != &p->mnt_mounts) {
221 p = list_entry(prev, struct vfsmount, mnt_child);
222 prev = p->mnt_mounts.prev;
223 }
224 return p;
225}
226
Ram Pai36341f62005-11-07 17:17:22 -0500227static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
228 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229{
230 struct super_block *sb = old->mnt_sb;
231 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
232
233 if (mnt) {
234 mnt->mnt_flags = old->mnt_flags;
235 atomic_inc(&sb->s_active);
236 mnt->mnt_sb = sb;
237 mnt->mnt_root = dget(root);
238 mnt->mnt_mountpoint = mnt->mnt_root;
239 mnt->mnt_parent = mnt;
Ram Paib90fa9a2005-11-07 17:19:50 -0500240
Ram Pai5afe0022005-11-07 17:21:01 -0500241 if (flag & CL_SLAVE) {
242 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
243 mnt->mnt_master = old;
244 CLEAR_MNT_SHARED(mnt);
245 } else {
246 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
247 list_add(&mnt->mnt_share, &old->mnt_share);
248 if (IS_MNT_SLAVE(old))
249 list_add(&mnt->mnt_slave, &old->mnt_slave);
250 mnt->mnt_master = old->mnt_master;
251 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500252 if (flag & CL_MAKE_SHARED)
253 set_mnt_shared(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254
255 /* stick the duplicate mount on the same expiry list
256 * as the original if that was on one */
Ram Pai36341f62005-11-07 17:17:22 -0500257 if (flag & CL_EXPIRE) {
258 spin_lock(&vfsmount_lock);
259 if (!list_empty(&old->mnt_expire))
260 list_add(&mnt->mnt_expire, &old->mnt_expire);
261 spin_unlock(&vfsmount_lock);
262 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 }
264 return mnt;
265}
266
Al Viro7b7b1ac2005-11-07 17:13:39 -0500267static inline void __mntput(struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268{
269 struct super_block *sb = mnt->mnt_sb;
270 dput(mnt->mnt_root);
271 free_vfsmnt(mnt);
272 deactivate_super(sb);
273}
274
Al Viro7b7b1ac2005-11-07 17:13:39 -0500275void mntput_no_expire(struct vfsmount *mnt)
276{
277repeat:
278 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
279 if (likely(!mnt->mnt_pinned)) {
280 spin_unlock(&vfsmount_lock);
281 __mntput(mnt);
282 return;
283 }
284 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
285 mnt->mnt_pinned = 0;
286 spin_unlock(&vfsmount_lock);
287 acct_auto_close_mnt(mnt);
288 security_sb_umount_close(mnt);
289 goto repeat;
290 }
291}
292
293EXPORT_SYMBOL(mntput_no_expire);
294
295void mnt_pin(struct vfsmount *mnt)
296{
297 spin_lock(&vfsmount_lock);
298 mnt->mnt_pinned++;
299 spin_unlock(&vfsmount_lock);
300}
301
302EXPORT_SYMBOL(mnt_pin);
303
304void mnt_unpin(struct vfsmount *mnt)
305{
306 spin_lock(&vfsmount_lock);
307 if (mnt->mnt_pinned) {
308 atomic_inc(&mnt->mnt_count);
309 mnt->mnt_pinned--;
310 }
311 spin_unlock(&vfsmount_lock);
312}
313
314EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315
316/* iterator */
317static void *m_start(struct seq_file *m, loff_t *pos)
318{
319 struct namespace *n = m->private;
320 struct list_head *p;
321 loff_t l = *pos;
322
Ram Pai390c6842005-11-07 17:17:51 -0500323 down_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 list_for_each(p, &n->list)
325 if (!l--)
326 return list_entry(p, struct vfsmount, mnt_list);
327 return NULL;
328}
329
330static void *m_next(struct seq_file *m, void *v, loff_t *pos)
331{
332 struct namespace *n = m->private;
333 struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
334 (*pos)++;
Ram Paib58fed82005-11-07 17:16:09 -0500335 return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336}
337
338static void m_stop(struct seq_file *m, void *v)
339{
Ram Pai390c6842005-11-07 17:17:51 -0500340 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341}
342
343static inline void mangle(struct seq_file *m, const char *s)
344{
345 seq_escape(m, s, " \t\n\\");
346}
347
348static int show_vfsmnt(struct seq_file *m, void *v)
349{
350 struct vfsmount *mnt = v;
351 int err = 0;
352 static struct proc_fs_info {
353 int flag;
354 char *str;
355 } fs_info[] = {
356 { MS_SYNCHRONOUS, ",sync" },
357 { MS_DIRSYNC, ",dirsync" },
358 { MS_MANDLOCK, ",mand" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 { 0, NULL }
360 };
361 static struct proc_fs_info mnt_info[] = {
362 { MNT_NOSUID, ",nosuid" },
363 { MNT_NODEV, ",nodev" },
364 { MNT_NOEXEC, ",noexec" },
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -0800365 { MNT_NOATIME, ",noatime" },
366 { MNT_NODIRATIME, ",nodiratime" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 { 0, NULL }
368 };
369 struct proc_fs_info *fs_infop;
370
371 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
372 seq_putc(m, ' ');
373 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
374 seq_putc(m, ' ');
375 mangle(m, mnt->mnt_sb->s_type->name);
376 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
377 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
378 if (mnt->mnt_sb->s_flags & fs_infop->flag)
379 seq_puts(m, fs_infop->str);
380 }
381 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
382 if (mnt->mnt_flags & fs_infop->flag)
383 seq_puts(m, fs_infop->str);
384 }
385 if (mnt->mnt_sb->s_op->show_options)
386 err = mnt->mnt_sb->s_op->show_options(m, mnt);
387 seq_puts(m, " 0 0\n");
388 return err;
389}
390
391struct seq_operations mounts_op = {
392 .start = m_start,
393 .next = m_next,
394 .stop = m_stop,
395 .show = show_vfsmnt
396};
397
398/**
399 * may_umount_tree - check if a mount tree is busy
400 * @mnt: root of mount tree
401 *
402 * This is called to check if a tree of mounts has any
403 * open files, pwds, chroots or sub mounts that are
404 * busy.
405 */
406int may_umount_tree(struct vfsmount *mnt)
407{
Ram Pai36341f62005-11-07 17:17:22 -0500408 int actual_refs = 0;
409 int minimum_refs = 0;
410 struct vfsmount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
412 spin_lock(&vfsmount_lock);
Ram Pai36341f62005-11-07 17:17:22 -0500413 for (p = mnt; p; p = next_mnt(p, mnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 actual_refs += atomic_read(&p->mnt_count);
415 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 }
417 spin_unlock(&vfsmount_lock);
418
419 if (actual_refs > minimum_refs)
420 return -EBUSY;
421
422 return 0;
423}
424
425EXPORT_SYMBOL(may_umount_tree);
426
427/**
428 * may_umount - check if a mount point is busy
429 * @mnt: root of mount
430 *
431 * This is called to check if a mount point has any
432 * open files, pwds, chroots or sub mounts. If the
433 * mount has sub mounts this will return busy
434 * regardless of whether the sub mounts are busy.
435 *
436 * Doesn't take quota and stuff into account. IOW, in some cases it will
437 * give false negatives. The main reason why it's here is that we need
438 * a non-destructive way to look for easily umountable filesystems.
439 */
440int may_umount(struct vfsmount *mnt)
441{
Ram Paia05964f2005-11-07 17:20:17 -0500442 int ret = 0;
443 spin_lock(&vfsmount_lock);
444 if (propagate_mount_busy(mnt, 2))
445 ret = -EBUSY;
446 spin_unlock(&vfsmount_lock);
447 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448}
449
450EXPORT_SYMBOL(may_umount);
451
Ram Paib90fa9a2005-11-07 17:19:50 -0500452void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
Ram Pai70fbcdf2005-11-07 17:17:04 -0500454 struct vfsmount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -0800455 while (!list_empty(head)) {
Ram Pai70fbcdf2005-11-07 17:17:04 -0500456 mnt = list_entry(head->next, struct vfsmount, mnt_hash);
457 list_del_init(&mnt->mnt_hash);
458 if (mnt->mnt_parent != mnt) {
459 struct dentry *dentry;
460 struct vfsmount *m;
461 spin_lock(&vfsmount_lock);
462 dentry = mnt->mnt_mountpoint;
463 m = mnt->mnt_parent;
464 mnt->mnt_mountpoint = mnt->mnt_root;
465 mnt->mnt_parent = mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500467 dput(dentry);
468 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 }
470 mntput(mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500471 }
472}
473
Ram Paia05964f2005-11-07 17:20:17 -0500474void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -0500475{
476 struct vfsmount *p;
477
478 for (p = mnt; p; p = next_mnt(p, mnt)) {
479 list_del(&p->mnt_hash);
480 list_add(&p->mnt_hash, kill);
481 }
482
Ram Paia05964f2005-11-07 17:20:17 -0500483 if (propagate)
484 propagate_umount(kill);
485
Ram Pai70fbcdf2005-11-07 17:17:04 -0500486 list_for_each_entry(p, kill, mnt_hash) {
487 list_del_init(&p->mnt_expire);
488 list_del_init(&p->mnt_list);
489 __touch_namespace(p->mnt_namespace);
490 p->mnt_namespace = NULL;
491 list_del_init(&p->mnt_child);
492 if (p->mnt_parent != p)
493 mnt->mnt_mountpoint->d_mounted--;
Ram Paia05964f2005-11-07 17:20:17 -0500494 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 }
496}
497
498static int do_umount(struct vfsmount *mnt, int flags)
499{
Ram Paib58fed82005-11-07 17:16:09 -0500500 struct super_block *sb = mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -0500502 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503
504 retval = security_sb_umount(mnt, flags);
505 if (retval)
506 return retval;
507
508 /*
509 * Allow userspace to request a mountpoint be expired rather than
510 * unmounting unconditionally. Unmount only happens if:
511 * (1) the mark is already set (the mark is cleared by mntput())
512 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
513 */
514 if (flags & MNT_EXPIRE) {
515 if (mnt == current->fs->rootmnt ||
516 flags & (MNT_FORCE | MNT_DETACH))
517 return -EINVAL;
518
519 if (atomic_read(&mnt->mnt_count) != 2)
520 return -EBUSY;
521
522 if (!xchg(&mnt->mnt_expiry_mark, 1))
523 return -EAGAIN;
524 }
525
526 /*
527 * If we may have to abort operations to get out of this
528 * mount, and they will themselves hold resources we must
529 * allow the fs to do things. In the Unix tradition of
530 * 'Gee thats tricky lets do it in userspace' the umount_begin
531 * might fail to complete on the first run through as other tasks
532 * must return, and the like. Thats for the mount program to worry
533 * about for the moment.
534 */
535
536 lock_kernel();
Ram Paib58fed82005-11-07 17:16:09 -0500537 if ((flags & MNT_FORCE) && sb->s_op->umount_begin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 sb->s_op->umount_begin(sb);
539 unlock_kernel();
540
541 /*
542 * No sense to grab the lock for this test, but test itself looks
543 * somewhat bogus. Suggestions for better replacement?
544 * Ho-hum... In principle, we might treat that as umount + switch
545 * to rootfs. GC would eventually take care of the old vfsmount.
546 * Actually it makes sense, especially if rootfs would contain a
547 * /reboot - static binary that would close all descriptors and
548 * call reboot(9). Then init(8) could umount root and exec /reboot.
549 */
550 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
551 /*
552 * Special case for "unmounting" root ...
553 * we just try to remount it readonly.
554 */
555 down_write(&sb->s_umount);
556 if (!(sb->s_flags & MS_RDONLY)) {
557 lock_kernel();
558 DQUOT_OFF(sb);
559 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
560 unlock_kernel();
561 }
562 up_write(&sb->s_umount);
563 return retval;
564 }
565
Ram Pai390c6842005-11-07 17:17:51 -0500566 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 spin_lock(&vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -0500568 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -0500571 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 if (!list_empty(&mnt->mnt_list))
Ram Paia05964f2005-11-07 17:20:17 -0500573 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 retval = 0;
575 }
576 spin_unlock(&vfsmount_lock);
577 if (retval)
578 security_sb_umount_busy(mnt);
Ram Pai390c6842005-11-07 17:17:51 -0500579 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500580 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 return retval;
582}
583
584/*
585 * Now umount can handle mount points as well as block devices.
586 * This is important for filesystems which use unnamed block devices.
587 *
588 * We now support a flag for forced unmount like the other 'big iron'
589 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
590 */
591
592asmlinkage long sys_umount(char __user * name, int flags)
593{
594 struct nameidata nd;
595 int retval;
596
597 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
598 if (retval)
599 goto out;
600 retval = -EINVAL;
601 if (nd.dentry != nd.mnt->mnt_root)
602 goto dput_and_out;
603 if (!check_mnt(nd.mnt))
604 goto dput_and_out;
605
606 retval = -EPERM;
607 if (!capable(CAP_SYS_ADMIN))
608 goto dput_and_out;
609
610 retval = do_umount(nd.mnt, flags);
611dput_and_out:
612 path_release_on_umount(&nd);
613out:
614 return retval;
615}
616
617#ifdef __ARCH_WANT_SYS_OLDUMOUNT
618
619/*
Ram Paib58fed82005-11-07 17:16:09 -0500620 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622asmlinkage long sys_oldumount(char __user * name)
623{
Ram Paib58fed82005-11-07 17:16:09 -0500624 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625}
626
627#endif
628
629static int mount_is_safe(struct nameidata *nd)
630{
631 if (capable(CAP_SYS_ADMIN))
632 return 0;
633 return -EPERM;
634#ifdef notyet
635 if (S_ISLNK(nd->dentry->d_inode->i_mode))
636 return -EPERM;
637 if (nd->dentry->d_inode->i_mode & S_ISVTX) {
638 if (current->uid != nd->dentry->d_inode->i_uid)
639 return -EPERM;
640 }
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800641 if (vfs_permission(nd, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 return -EPERM;
643 return 0;
644#endif
645}
646
Ram Paib58fed82005-11-07 17:16:09 -0500647static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648{
649 while (1) {
650 if (d == dentry)
651 return 1;
652 if (d == NULL || d == d->d_parent)
653 return 0;
654 d = d->d_parent;
655 }
656}
657
Ram Paib90fa9a2005-11-07 17:19:50 -0500658struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -0500659 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660{
661 struct vfsmount *res, *p, *q, *r, *s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 struct nameidata nd;
663
Ram Pai9676f0c2005-11-07 17:21:20 -0500664 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
665 return NULL;
666
Ram Pai36341f62005-11-07 17:17:22 -0500667 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 if (!q)
669 goto Enomem;
670 q->mnt_mountpoint = mnt->mnt_mountpoint;
671
672 p = mnt;
Domen Puncerfdadd652005-09-10 00:27:07 -0700673 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
675 continue;
676
677 for (s = r; s; s = next_mnt(s, r)) {
Ram Pai9676f0c2005-11-07 17:21:20 -0500678 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
679 s = skip_mnt_tree(s);
680 continue;
681 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 while (p != s->mnt_parent) {
683 p = p->mnt_parent;
684 q = q->mnt_parent;
685 }
686 p = s;
687 nd.mnt = q;
688 nd.dentry = p->mnt_mountpoint;
Ram Pai36341f62005-11-07 17:17:22 -0500689 q = clone_mnt(p, p->mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 if (!q)
691 goto Enomem;
692 spin_lock(&vfsmount_lock);
693 list_add_tail(&q->mnt_list, &res->mnt_list);
694 attach_mnt(q, &nd);
695 spin_unlock(&vfsmount_lock);
696 }
697 }
698 return res;
Ram Paib58fed82005-11-07 17:16:09 -0500699Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -0500701 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -0500703 umount_tree(res, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500705 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 }
707 return NULL;
708}
709
Ram Paib90fa9a2005-11-07 17:19:50 -0500710/*
711 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -0500712 * @nd : place the mount tree @source_mnt is attached
713 * @parent_nd : if non-null, detach the source_mnt from its parent and
714 * store the parent mount and mountpoint dentry.
715 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -0500716 *
717 * NOTE: in the table below explains the semantics when a source mount
718 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -0500719 * ---------------------------------------------------------------------------
720 * | BIND MOUNT OPERATION |
721 * |**************************************************************************
722 * | source-->| shared | private | slave | unbindable |
723 * | dest | | | | |
724 * | | | | | | |
725 * | v | | | | |
726 * |**************************************************************************
727 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
728 * | | | | | |
729 * |non-shared| shared (+) | private | slave (*) | invalid |
730 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -0500731 * A bind operation clones the source mount and mounts the clone on the
732 * destination mount.
733 *
734 * (++) the cloned mount is propagated to all the mounts in the propagation
735 * tree of the destination mount and the cloned mount is added to
736 * the peer group of the source mount.
737 * (+) the cloned mount is created under the destination mount and is marked
738 * as shared. The cloned mount is added to the peer group of the source
739 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -0500740 * (+++) the mount is propagated to all the mounts in the propagation tree
741 * of the destination mount and the cloned mount is made slave
742 * of the same master as that of the source mount. The cloned mount
743 * is marked as 'shared and slave'.
744 * (*) the cloned mount is made a slave of the same master as that of the
745 * source mount.
746 *
Ram Pai9676f0c2005-11-07 17:21:20 -0500747 * ---------------------------------------------------------------------------
748 * | MOVE MOUNT OPERATION |
749 * |**************************************************************************
750 * | source-->| shared | private | slave | unbindable |
751 * | dest | | | | |
752 * | | | | | | |
753 * | v | | | | |
754 * |**************************************************************************
755 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
756 * | | | | | |
757 * |non-shared| shared (+*) | private | slave (*) | unbindable |
758 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -0500759 *
760 * (+) the mount is moved to the destination. And is then propagated to
761 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -0500762 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -0500763 * (+++) the mount is moved to the destination and is then propagated to
764 * all the mounts belonging to the destination mount's propagation tree.
765 * the mount is marked as 'shared and slave'.
766 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -0500767 *
768 * if the source mount is a tree, the operations explained above is
769 * applied to each mount in the tree.
770 * Must be called without spinlocks held, since this function can sleep
771 * in allocations.
772 */
773static int attach_recursive_mnt(struct vfsmount *source_mnt,
Ram Pai21444402005-11-07 17:20:03 -0500774 struct nameidata *nd, struct nameidata *parent_nd)
Ram Paib90fa9a2005-11-07 17:19:50 -0500775{
776 LIST_HEAD(tree_list);
777 struct vfsmount *dest_mnt = nd->mnt;
778 struct dentry *dest_dentry = nd->dentry;
779 struct vfsmount *child, *p;
780
781 if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
782 return -EINVAL;
783
784 if (IS_MNT_SHARED(dest_mnt)) {
785 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
786 set_mnt_shared(p);
787 }
788
789 spin_lock(&vfsmount_lock);
Ram Pai21444402005-11-07 17:20:03 -0500790 if (parent_nd) {
791 detach_mnt(source_mnt, parent_nd);
792 attach_mnt(source_mnt, nd);
793 touch_namespace(current->namespace);
794 } else {
795 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
796 commit_tree(source_mnt);
797 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500798
799 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
800 list_del_init(&child->mnt_hash);
801 commit_tree(child);
802 }
803 spin_unlock(&vfsmount_lock);
804 return 0;
805}
806
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
808{
809 int err;
810 if (mnt->mnt_sb->s_flags & MS_NOUSER)
811 return -EINVAL;
812
813 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
814 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
815 return -ENOTDIR;
816
817 err = -ENOENT;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800818 mutex_lock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 if (IS_DEADDIR(nd->dentry->d_inode))
820 goto out_unlock;
821
822 err = security_sb_check_sb(mnt, nd);
823 if (err)
824 goto out_unlock;
825
826 err = -ENOENT;
Ram Paib90fa9a2005-11-07 17:19:50 -0500827 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry))
Ram Pai21444402005-11-07 17:20:03 -0500828 err = attach_recursive_mnt(mnt, nd, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829out_unlock:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800830 mutex_unlock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 if (!err)
832 security_sb_post_addmount(mnt, nd);
833 return err;
834}
835
836/*
Ram Pai07b20882005-11-07 17:19:07 -0500837 * recursively change the type of the mountpoint.
838 */
839static int do_change_type(struct nameidata *nd, int flag)
840{
841 struct vfsmount *m, *mnt = nd->mnt;
842 int recurse = flag & MS_REC;
843 int type = flag & ~MS_REC;
844
845 if (nd->dentry != nd->mnt->mnt_root)
846 return -EINVAL;
847
848 down_write(&namespace_sem);
849 spin_lock(&vfsmount_lock);
850 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
851 change_mnt_propagation(m, type);
852 spin_unlock(&vfsmount_lock);
853 up_write(&namespace_sem);
854 return 0;
855}
856
857/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 * do loopback mount.
859 */
860static int do_loopback(struct nameidata *nd, char *old_name, int recurse)
861{
862 struct nameidata old_nd;
863 struct vfsmount *mnt = NULL;
864 int err = mount_is_safe(nd);
865 if (err)
866 return err;
867 if (!old_name || !*old_name)
868 return -EINVAL;
869 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
870 if (err)
871 return err;
872
Ram Pai390c6842005-11-07 17:17:51 -0500873 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 err = -EINVAL;
Ram Pai9676f0c2005-11-07 17:21:20 -0500875 if (IS_MNT_UNBINDABLE(old_nd.mnt))
876 goto out;
877
Al Viroccd48bc2005-11-07 17:15:04 -0500878 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
879 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880
Al Viroccd48bc2005-11-07 17:15:04 -0500881 err = -ENOMEM;
882 if (recurse)
Ram Pai36341f62005-11-07 17:17:22 -0500883 mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -0500884 else
Ram Pai36341f62005-11-07 17:17:22 -0500885 mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -0500886
887 if (!mnt)
888 goto out;
889
Al Viroccd48bc2005-11-07 17:15:04 -0500890 err = graft_tree(mnt, nd);
891 if (err) {
Ram Pai70fbcdf2005-11-07 17:17:04 -0500892 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -0500894 umount_tree(mnt, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500896 release_mounts(&umount_list);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -0500897 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898
Al Viroccd48bc2005-11-07 17:15:04 -0500899out:
Ram Pai390c6842005-11-07 17:17:51 -0500900 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 path_release(&old_nd);
902 return err;
903}
904
905/*
906 * change filesystem flags. dir should be a physical root of filesystem.
907 * If you've mounted a non-root directory somewhere and want to do remount
908 * on it - tough luck.
909 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
911 void *data)
912{
913 int err;
Ram Paib58fed82005-11-07 17:16:09 -0500914 struct super_block *sb = nd->mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915
916 if (!capable(CAP_SYS_ADMIN))
917 return -EPERM;
918
919 if (!check_mnt(nd->mnt))
920 return -EINVAL;
921
922 if (nd->dentry != nd->mnt->mnt_root)
923 return -EINVAL;
924
925 down_write(&sb->s_umount);
926 err = do_remount_sb(sb, flags, data, 0);
927 if (!err)
Ram Paib58fed82005-11-07 17:16:09 -0500928 nd->mnt->mnt_flags = mnt_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 up_write(&sb->s_umount);
930 if (!err)
931 security_sb_post_remount(nd->mnt, flags, data);
932 return err;
933}
934
Ram Pai9676f0c2005-11-07 17:21:20 -0500935static inline int tree_contains_unbindable(struct vfsmount *mnt)
936{
937 struct vfsmount *p;
938 for (p = mnt; p; p = next_mnt(p, mnt)) {
939 if (IS_MNT_UNBINDABLE(p))
940 return 1;
941 }
942 return 0;
943}
944
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945static int do_move_mount(struct nameidata *nd, char *old_name)
946{
947 struct nameidata old_nd, parent_nd;
948 struct vfsmount *p;
949 int err = 0;
950 if (!capable(CAP_SYS_ADMIN))
951 return -EPERM;
952 if (!old_name || !*old_name)
953 return -EINVAL;
954 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
955 if (err)
956 return err;
957
Ram Pai390c6842005-11-07 17:17:51 -0500958 down_write(&namespace_sem);
Ram Paib58fed82005-11-07 17:16:09 -0500959 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 ;
961 err = -EINVAL;
962 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
963 goto out;
964
965 err = -ENOENT;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800966 mutex_lock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 if (IS_DEADDIR(nd->dentry->d_inode))
968 goto out1;
969
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
Ram Pai21444402005-11-07 17:20:03 -0500971 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
973 err = -EINVAL;
974 if (old_nd.dentry != old_nd.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -0500975 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976
977 if (old_nd.mnt == old_nd.mnt->mnt_parent)
Ram Pai21444402005-11-07 17:20:03 -0500978 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979
980 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
981 S_ISDIR(old_nd.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -0500982 goto out1;
983 /*
984 * Don't move a mount residing in a shared parent.
985 */
986 if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent))
987 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -0500988 /*
989 * Don't move a mount tree containing unbindable mounts to a destination
990 * mount which is shared.
991 */
992 if (IS_MNT_SHARED(nd->mnt) && tree_contains_unbindable(old_nd.mnt))
993 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 err = -ELOOP;
Ram Paib58fed82005-11-07 17:16:09 -0500995 for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 if (p == old_nd.mnt)
Ram Pai21444402005-11-07 17:20:03 -0500997 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998
Ram Pai21444402005-11-07 17:20:03 -0500999 if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd)))
1000 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001
Ram Pai21444402005-11-07 17:20:03 -05001002 spin_lock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 /* if the mount is moved, it should no longer be expire
1004 * automatically */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001005 list_del_init(&old_nd.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 spin_unlock(&vfsmount_lock);
1007out1:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001008 mutex_unlock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009out:
Ram Pai390c6842005-11-07 17:17:51 -05001010 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 if (!err)
1012 path_release(&parent_nd);
1013 path_release(&old_nd);
1014 return err;
1015}
1016
1017/*
1018 * create a new mount for userspace and request it to be added into the
1019 * namespace's tree
1020 */
1021static int do_new_mount(struct nameidata *nd, char *type, int flags,
1022 int mnt_flags, char *name, void *data)
1023{
1024 struct vfsmount *mnt;
1025
1026 if (!type || !memchr(type, 0, PAGE_SIZE))
1027 return -EINVAL;
1028
1029 /* we need capabilities... */
1030 if (!capable(CAP_SYS_ADMIN))
1031 return -EPERM;
1032
1033 mnt = do_kern_mount(type, flags, name, data);
1034 if (IS_ERR(mnt))
1035 return PTR_ERR(mnt);
1036
1037 return do_add_mount(mnt, nd, mnt_flags, NULL);
1038}
1039
1040/*
1041 * add a mount into a namespace's mount tree
1042 * - provide the option of adding the new mount to an expiration list
1043 */
1044int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1045 int mnt_flags, struct list_head *fslist)
1046{
1047 int err;
1048
Ram Pai390c6842005-11-07 17:17:51 -05001049 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 /* Something was mounted here while we slept */
Ram Paib58fed82005-11-07 17:16:09 -05001051 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 ;
1053 err = -EINVAL;
1054 if (!check_mnt(nd->mnt))
1055 goto unlock;
1056
1057 /* Refuse the same filesystem on the same mount point */
1058 err = -EBUSY;
1059 if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
1060 nd->mnt->mnt_root == nd->dentry)
1061 goto unlock;
1062
1063 err = -EINVAL;
1064 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1065 goto unlock;
1066
1067 newmnt->mnt_flags = mnt_flags;
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001068 if ((err = graft_tree(newmnt, nd)))
1069 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001071 if (fslist) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 /* add to the specified expiration list */
1073 spin_lock(&vfsmount_lock);
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001074 list_add_tail(&newmnt->mnt_expire, fslist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 spin_unlock(&vfsmount_lock);
1076 }
Ram Pai390c6842005-11-07 17:17:51 -05001077 up_write(&namespace_sem);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001078 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079
1080unlock:
Ram Pai390c6842005-11-07 17:17:51 -05001081 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 mntput(newmnt);
1083 return err;
1084}
1085
1086EXPORT_SYMBOL_GPL(do_add_mount);
1087
Ram Pai70fbcdf2005-11-07 17:17:04 -05001088static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
1089 struct list_head *umounts)
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001090{
1091 spin_lock(&vfsmount_lock);
1092
1093 /*
Miklos Szeredied42c872005-07-07 17:57:26 -07001094 * Check if mount is still attached, if not, let whoever holds it deal
1095 * with the sucker
1096 */
1097 if (mnt->mnt_parent == mnt) {
1098 spin_unlock(&vfsmount_lock);
1099 return;
1100 }
1101
1102 /*
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001103 * Check that it is still dead: the count should now be 2 - as
1104 * contributed by the vfsmount parent and the mntget above
1105 */
Ram Paia05964f2005-11-07 17:20:17 -05001106 if (!propagate_mount_busy(mnt, 2)) {
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001107 /* delete from the namespace */
Al Viro5addc5d2005-11-07 17:15:49 -05001108 touch_namespace(mnt->mnt_namespace);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001109 list_del_init(&mnt->mnt_list);
Miklos Szerediac081152005-07-07 17:57:27 -07001110 mnt->mnt_namespace = NULL;
Ram Paia05964f2005-11-07 17:20:17 -05001111 umount_tree(mnt, 1, umounts);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001112 spin_unlock(&vfsmount_lock);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001113 } else {
1114 /*
1115 * Someone brought it back to life whilst we didn't have any
1116 * locks held so return it to the expiration list
1117 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001118 list_add_tail(&mnt->mnt_expire, mounts);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001119 spin_unlock(&vfsmount_lock);
1120 }
1121}
1122
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123/*
1124 * process a list of expirable mountpoints with the intent of discarding any
1125 * mountpoints that aren't in use and haven't been touched since last we came
1126 * here
1127 */
1128void mark_mounts_for_expiry(struct list_head *mounts)
1129{
1130 struct namespace *namespace;
1131 struct vfsmount *mnt, *next;
1132 LIST_HEAD(graveyard);
1133
1134 if (list_empty(mounts))
1135 return;
1136
1137 spin_lock(&vfsmount_lock);
1138
1139 /* extract from the expiration list every vfsmount that matches the
1140 * following criteria:
1141 * - only referenced by its parent vfsmount
1142 * - still marked for expiry (marked on the last call here; marks are
1143 * cleared by mntput())
1144 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001145 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1147 atomic_read(&mnt->mnt_count) != 1)
1148 continue;
1149
1150 mntget(mnt);
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001151 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 }
1153
1154 /*
1155 * go through the vfsmounts we've just consigned to the graveyard to
1156 * - check that they're still dead
1157 * - delete the vfsmount from the appropriate namespace under lock
1158 * - dispose of the corpse
1159 */
1160 while (!list_empty(&graveyard)) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001161 LIST_HEAD(umounts);
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001162 mnt = list_entry(graveyard.next, struct vfsmount, mnt_expire);
1163 list_del_init(&mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164
1165 /* don't do anything if the namespace is dead - all the
1166 * vfsmounts from it are going away anyway */
1167 namespace = mnt->mnt_namespace;
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -07001168 if (!namespace || !namespace->root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 continue;
1170 get_namespace(namespace);
1171
1172 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001173 down_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001174 expire_mount(mnt, mounts, &umounts);
Ram Pai390c6842005-11-07 17:17:51 -05001175 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001176 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 mntput(mnt);
1178 put_namespace(namespace);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 spin_lock(&vfsmount_lock);
1180 }
1181
1182 spin_unlock(&vfsmount_lock);
1183}
1184
1185EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1186
1187/*
1188 * Some copy_from_user() implementations do not return the exact number of
1189 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1190 * Note that this function differs from copy_from_user() in that it will oops
1191 * on bad values of `to', rather than returning a short copy.
1192 */
Ram Paib58fed82005-11-07 17:16:09 -05001193static long exact_copy_from_user(void *to, const void __user * from,
1194 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195{
1196 char *t = to;
1197 const char __user *f = from;
1198 char c;
1199
1200 if (!access_ok(VERIFY_READ, from, n))
1201 return n;
1202
1203 while (n) {
1204 if (__get_user(c, f)) {
1205 memset(t, 0, n);
1206 break;
1207 }
1208 *t++ = c;
1209 f++;
1210 n--;
1211 }
1212 return n;
1213}
1214
Ram Paib58fed82005-11-07 17:16:09 -05001215int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216{
1217 int i;
1218 unsigned long page;
1219 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05001220
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 *where = 0;
1222 if (!data)
1223 return 0;
1224
1225 if (!(page = __get_free_page(GFP_KERNEL)))
1226 return -ENOMEM;
1227
1228 /* We only care that *some* data at the address the user
1229 * gave us is valid. Just in case, we'll zero
1230 * the remainder of the page.
1231 */
1232 /* copy_from_user cannot cross TASK_SIZE ! */
1233 size = TASK_SIZE - (unsigned long)data;
1234 if (size > PAGE_SIZE)
1235 size = PAGE_SIZE;
1236
1237 i = size - exact_copy_from_user((void *)page, data, size);
1238 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05001239 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 return -EFAULT;
1241 }
1242 if (i != PAGE_SIZE)
1243 memset((char *)page + i, 0, PAGE_SIZE - i);
1244 *where = page;
1245 return 0;
1246}
1247
1248/*
1249 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1250 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1251 *
1252 * data is a (void *) that can point to any structure up to
1253 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1254 * information (or be NULL).
1255 *
1256 * Pre-0.97 versions of mount() didn't have a flags word.
1257 * When the flags word was introduced its top half was required
1258 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1259 * Therefore, if this magic number is present, it carries no information
1260 * and must be discarded.
1261 */
Ram Paib58fed82005-11-07 17:16:09 -05001262long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 unsigned long flags, void *data_page)
1264{
1265 struct nameidata nd;
1266 int retval = 0;
1267 int mnt_flags = 0;
1268
1269 /* Discard magic */
1270 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1271 flags &= ~MS_MGC_MSK;
1272
1273 /* Basic sanity checks */
1274
1275 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1276 return -EINVAL;
1277 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1278 return -EINVAL;
1279
1280 if (data_page)
1281 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1282
1283 /* Separate the per-mountpoint flags */
1284 if (flags & MS_NOSUID)
1285 mnt_flags |= MNT_NOSUID;
1286 if (flags & MS_NODEV)
1287 mnt_flags |= MNT_NODEV;
1288 if (flags & MS_NOEXEC)
1289 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001290 if (flags & MS_NOATIME)
1291 mnt_flags |= MNT_NOATIME;
1292 if (flags & MS_NODIRATIME)
1293 mnt_flags |= MNT_NODIRATIME;
1294
1295 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
1296 MS_NOATIME | MS_NODIRATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297
1298 /* ... and get the mountpoint */
1299 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1300 if (retval)
1301 return retval;
1302
1303 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1304 if (retval)
1305 goto dput_out;
1306
1307 if (flags & MS_REMOUNT)
1308 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1309 data_page);
1310 else if (flags & MS_BIND)
1311 retval = do_loopback(&nd, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05001312 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Ram Pai07b20882005-11-07 17:19:07 -05001313 retval = do_change_type(&nd, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 else if (flags & MS_MOVE)
1315 retval = do_move_mount(&nd, dev_name);
1316 else
1317 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1318 dev_name, data_page);
1319dput_out:
1320 path_release(&nd);
1321 return retval;
1322}
1323
1324int copy_namespace(int flags, struct task_struct *tsk)
1325{
1326 struct namespace *namespace = tsk->namespace;
1327 struct namespace *new_ns;
1328 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
1329 struct fs_struct *fs = tsk->fs;
1330 struct vfsmount *p, *q;
1331
1332 if (!namespace)
1333 return 0;
1334
1335 get_namespace(namespace);
1336
1337 if (!(flags & CLONE_NEWNS))
1338 return 0;
1339
1340 if (!capable(CAP_SYS_ADMIN)) {
1341 put_namespace(namespace);
1342 return -EPERM;
1343 }
1344
1345 new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL);
1346 if (!new_ns)
1347 goto out;
1348
1349 atomic_set(&new_ns->count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 INIT_LIST_HEAD(&new_ns->list);
Al Viro5addc5d2005-11-07 17:15:49 -05001351 init_waitqueue_head(&new_ns->poll);
1352 new_ns->event = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353
Ram Pai390c6842005-11-07 17:17:51 -05001354 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 /* First pass: copy the tree topology */
Ram Pai36341f62005-11-07 17:17:22 -05001356 new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05001357 CL_COPY_ALL | CL_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 if (!new_ns->root) {
Ram Pai390c6842005-11-07 17:17:51 -05001359 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 kfree(new_ns);
1361 goto out;
1362 }
1363 spin_lock(&vfsmount_lock);
1364 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1365 spin_unlock(&vfsmount_lock);
1366
1367 /*
1368 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1369 * as belonging to new namespace. We have already acquired a private
1370 * fs_struct, so tsk->fs->lock is not needed.
1371 */
1372 p = namespace->root;
1373 q = new_ns->root;
1374 while (p) {
1375 q->mnt_namespace = new_ns;
1376 if (fs) {
1377 if (p == fs->rootmnt) {
1378 rootmnt = p;
1379 fs->rootmnt = mntget(q);
1380 }
1381 if (p == fs->pwdmnt) {
1382 pwdmnt = p;
1383 fs->pwdmnt = mntget(q);
1384 }
1385 if (p == fs->altrootmnt) {
1386 altrootmnt = p;
1387 fs->altrootmnt = mntget(q);
1388 }
1389 }
1390 p = next_mnt(p, namespace->root);
1391 q = next_mnt(q, new_ns->root);
1392 }
Ram Pai390c6842005-11-07 17:17:51 -05001393 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394
1395 tsk->namespace = new_ns;
1396
1397 if (rootmnt)
1398 mntput(rootmnt);
1399 if (pwdmnt)
1400 mntput(pwdmnt);
1401 if (altrootmnt)
1402 mntput(altrootmnt);
1403
1404 put_namespace(namespace);
1405 return 0;
1406
1407out:
1408 put_namespace(namespace);
1409 return -ENOMEM;
1410}
1411
1412asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1413 char __user * type, unsigned long flags,
1414 void __user * data)
1415{
1416 int retval;
1417 unsigned long data_page;
1418 unsigned long type_page;
1419 unsigned long dev_page;
1420 char *dir_page;
1421
Ram Paib58fed82005-11-07 17:16:09 -05001422 retval = copy_mount_options(type, &type_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 if (retval < 0)
1424 return retval;
1425
1426 dir_page = getname(dir_name);
1427 retval = PTR_ERR(dir_page);
1428 if (IS_ERR(dir_page))
1429 goto out1;
1430
Ram Paib58fed82005-11-07 17:16:09 -05001431 retval = copy_mount_options(dev_name, &dev_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 if (retval < 0)
1433 goto out2;
1434
Ram Paib58fed82005-11-07 17:16:09 -05001435 retval = copy_mount_options(data, &data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 if (retval < 0)
1437 goto out3;
1438
1439 lock_kernel();
Ram Paib58fed82005-11-07 17:16:09 -05001440 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1441 flags, (void *)data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 unlock_kernel();
1443 free_page(data_page);
1444
1445out3:
1446 free_page(dev_page);
1447out2:
1448 putname(dir_page);
1449out1:
1450 free_page(type_page);
1451 return retval;
1452}
1453
1454/*
1455 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1456 * It can block. Requires the big lock held.
1457 */
1458void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1459 struct dentry *dentry)
1460{
1461 struct dentry *old_root;
1462 struct vfsmount *old_rootmnt;
1463 write_lock(&fs->lock);
1464 old_root = fs->root;
1465 old_rootmnt = fs->rootmnt;
1466 fs->rootmnt = mntget(mnt);
1467 fs->root = dget(dentry);
1468 write_unlock(&fs->lock);
1469 if (old_root) {
1470 dput(old_root);
1471 mntput(old_rootmnt);
1472 }
1473}
1474
1475/*
1476 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1477 * It can block. Requires the big lock held.
1478 */
1479void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1480 struct dentry *dentry)
1481{
1482 struct dentry *old_pwd;
1483 struct vfsmount *old_pwdmnt;
1484
1485 write_lock(&fs->lock);
1486 old_pwd = fs->pwd;
1487 old_pwdmnt = fs->pwdmnt;
1488 fs->pwdmnt = mntget(mnt);
1489 fs->pwd = dget(dentry);
1490 write_unlock(&fs->lock);
1491
1492 if (old_pwd) {
1493 dput(old_pwd);
1494 mntput(old_pwdmnt);
1495 }
1496}
1497
1498static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1499{
1500 struct task_struct *g, *p;
1501 struct fs_struct *fs;
1502
1503 read_lock(&tasklist_lock);
1504 do_each_thread(g, p) {
1505 task_lock(p);
1506 fs = p->fs;
1507 if (fs) {
1508 atomic_inc(&fs->count);
1509 task_unlock(p);
Ram Paib58fed82005-11-07 17:16:09 -05001510 if (fs->root == old_nd->dentry
1511 && fs->rootmnt == old_nd->mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
Ram Paib58fed82005-11-07 17:16:09 -05001513 if (fs->pwd == old_nd->dentry
1514 && fs->pwdmnt == old_nd->mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1516 put_fs_struct(fs);
1517 } else
1518 task_unlock(p);
1519 } while_each_thread(g, p);
1520 read_unlock(&tasklist_lock);
1521}
1522
1523/*
1524 * pivot_root Semantics:
1525 * Moves the root file system of the current process to the directory put_old,
1526 * makes new_root as the new root file system of the current process, and sets
1527 * root/cwd of all processes which had them on the current root to new_root.
1528 *
1529 * Restrictions:
1530 * The new_root and put_old must be directories, and must not be on the
1531 * same file system as the current process root. The put_old must be
1532 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1533 * pointed to by put_old must yield the same directory as new_root. No other
1534 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1535 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08001536 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1537 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1538 * in this situation.
1539 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 * Notes:
1541 * - we don't move root/cwd if they are not at the root (reason: if something
1542 * cared enough to change them, it's probably wrong to force them elsewhere)
1543 * - it's okay to pick a root that isn't the root of a file system, e.g.
1544 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1545 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1546 * first.
1547 */
Ram Paib58fed82005-11-07 17:16:09 -05001548asmlinkage long sys_pivot_root(const char __user * new_root,
1549 const char __user * put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550{
1551 struct vfsmount *tmp;
1552 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1553 int error;
1554
1555 if (!capable(CAP_SYS_ADMIN))
1556 return -EPERM;
1557
1558 lock_kernel();
1559
Ram Paib58fed82005-11-07 17:16:09 -05001560 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1561 &new_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 if (error)
1563 goto out0;
1564 error = -EINVAL;
1565 if (!check_mnt(new_nd.mnt))
1566 goto out1;
1567
Ram Paib58fed82005-11-07 17:16:09 -05001568 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 if (error)
1570 goto out1;
1571
1572 error = security_sb_pivotroot(&old_nd, &new_nd);
1573 if (error) {
1574 path_release(&old_nd);
1575 goto out1;
1576 }
1577
1578 read_lock(&current->fs->lock);
1579 user_nd.mnt = mntget(current->fs->rootmnt);
1580 user_nd.dentry = dget(current->fs->root);
1581 read_unlock(&current->fs->lock);
Ram Pai390c6842005-11-07 17:17:51 -05001582 down_write(&namespace_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001583 mutex_lock(&old_nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 error = -EINVAL;
Ram Pai21444402005-11-07 17:20:03 -05001585 if (IS_MNT_SHARED(old_nd.mnt) ||
1586 IS_MNT_SHARED(new_nd.mnt->mnt_parent) ||
1587 IS_MNT_SHARED(user_nd.mnt->mnt_parent))
1588 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 if (!check_mnt(user_nd.mnt))
1590 goto out2;
1591 error = -ENOENT;
1592 if (IS_DEADDIR(new_nd.dentry->d_inode))
1593 goto out2;
1594 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1595 goto out2;
1596 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1597 goto out2;
1598 error = -EBUSY;
1599 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1600 goto out2; /* loop, on the same file system */
1601 error = -EINVAL;
1602 if (user_nd.mnt->mnt_root != user_nd.dentry)
1603 goto out2; /* not a mountpoint */
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07001604 if (user_nd.mnt->mnt_parent == user_nd.mnt)
1605 goto out2; /* not attached */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 if (new_nd.mnt->mnt_root != new_nd.dentry)
1607 goto out2; /* not a mountpoint */
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07001608 if (new_nd.mnt->mnt_parent == new_nd.mnt)
1609 goto out2; /* not attached */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1611 spin_lock(&vfsmount_lock);
1612 if (tmp != new_nd.mnt) {
1613 for (;;) {
1614 if (tmp->mnt_parent == tmp)
1615 goto out3; /* already mounted on put_old */
1616 if (tmp->mnt_parent == new_nd.mnt)
1617 break;
1618 tmp = tmp->mnt_parent;
1619 }
1620 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1621 goto out3;
1622 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1623 goto out3;
1624 detach_mnt(new_nd.mnt, &parent_nd);
1625 detach_mnt(user_nd.mnt, &root_parent);
1626 attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */
1627 attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
Al Viro5addc5d2005-11-07 17:15:49 -05001628 touch_namespace(current->namespace);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 spin_unlock(&vfsmount_lock);
1630 chroot_fs_refs(&user_nd, &new_nd);
1631 security_sb_post_pivotroot(&user_nd, &new_nd);
1632 error = 0;
1633 path_release(&root_parent);
1634 path_release(&parent_nd);
1635out2:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001636 mutex_unlock(&old_nd.dentry->d_inode->i_mutex);
Ram Pai390c6842005-11-07 17:17:51 -05001637 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 path_release(&user_nd);
1639 path_release(&old_nd);
1640out1:
1641 path_release(&new_nd);
1642out0:
1643 unlock_kernel();
1644 return error;
1645out3:
1646 spin_unlock(&vfsmount_lock);
1647 goto out2;
1648}
1649
1650static void __init init_mount_tree(void)
1651{
1652 struct vfsmount *mnt;
1653 struct namespace *namespace;
1654 struct task_struct *g, *p;
1655
1656 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1657 if (IS_ERR(mnt))
1658 panic("Can't create rootfs");
1659 namespace = kmalloc(sizeof(*namespace), GFP_KERNEL);
1660 if (!namespace)
1661 panic("Can't allocate initial namespace");
1662 atomic_set(&namespace->count, 1);
1663 INIT_LIST_HEAD(&namespace->list);
Al Viro5addc5d2005-11-07 17:15:49 -05001664 init_waitqueue_head(&namespace->poll);
1665 namespace->event = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 list_add(&mnt->mnt_list, &namespace->list);
1667 namespace->root = mnt;
1668 mnt->mnt_namespace = namespace;
1669
1670 init_task.namespace = namespace;
1671 read_lock(&tasklist_lock);
1672 do_each_thread(g, p) {
1673 get_namespace(namespace);
1674 p->namespace = namespace;
1675 } while_each_thread(g, p);
1676 read_unlock(&tasklist_lock);
1677
1678 set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root);
1679 set_fs_root(current->fs, namespace->root, namespace->root->mnt_root);
1680}
1681
1682void __init mnt_init(unsigned long mempages)
1683{
1684 struct list_head *d;
1685 unsigned int nr_hash;
1686 int i;
1687
Ram Pai390c6842005-11-07 17:17:51 -05001688 init_rwsem(&namespace_sem);
1689
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
Ram Paib58fed82005-11-07 17:16:09 -05001691 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692
Ram Paib58fed82005-11-07 17:16:09 -05001693 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694
1695 if (!mount_hashtable)
1696 panic("Failed to allocate mount hash table\n");
1697
1698 /*
1699 * Find the power-of-two list-heads that can fit into the allocation..
1700 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1701 * a power-of-two.
1702 */
1703 nr_hash = PAGE_SIZE / sizeof(struct list_head);
1704 hash_bits = 0;
1705 do {
1706 hash_bits++;
1707 } while ((nr_hash >> hash_bits) != 0);
1708 hash_bits--;
1709
1710 /*
1711 * Re-calculate the actual number of entries and the mask
1712 * from the number of bits we can fit.
1713 */
1714 nr_hash = 1UL << hash_bits;
Ram Paib58fed82005-11-07 17:16:09 -05001715 hash_mask = nr_hash - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716
1717 printk("Mount-cache hash table entries: %d\n", nr_hash);
1718
1719 /* And initialize the newly allocated array */
1720 d = mount_hashtable;
1721 i = nr_hash;
1722 do {
1723 INIT_LIST_HEAD(d);
1724 d++;
1725 i--;
1726 } while (i);
1727 sysfs_init();
1728 init_rootfs();
1729 init_mount_tree();
1730}
1731
1732void __put_namespace(struct namespace *namespace)
1733{
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -07001734 struct vfsmount *root = namespace->root;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001735 LIST_HEAD(umount_list);
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -07001736 namespace->root = NULL;
1737 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001738 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001740 umount_tree(root, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001742 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001743 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 kfree(namespace);
1745}